Labor Demand Shocks at Birth and Cognitive Achievement ...€¦ · 1E.g., UNICEF 2014, Woodhead...

Labor Demand Shocks at Birth and CognitiveAchievement during Childhood

Krishna Regmi∗ Daniel J. Henderson†

July 20, 2019

Abstract

As epidemiological studies have shown that conditions during gestation and earlychildhood affect adult health outcomes, we examine the effect of local labor marketconditions in the year of birth on cognitive development in childhood. To addressthe endogeneity of labor market conditions, we construct gender-specific predicted em-ployment growth rates at the state level by interacting an industry’s share in a state’semployment with the industry’s national growth rate. We find that an increase inemployment opportunities for men leads to an improvement in children’s cognitiveachievement as measured by reading and math test scores. Additionally, our estimatesshow a positive and significant effect of male-specific employment growth on children’sPeabody Picture Vocabulary Test scores and in home environment in the year of birth.We find an insignificant positive effect of buoyancy in females’ employment opportuni-ties on said test scores.

JEL classification: J20; J21; I20; I30Keywords: labor market conditions; cognitive ability; child’s well-being

∗Department of Ag. Economics and Economics, Montana State University†Department of Economics, Finance and Legal Studies, University of Alabama and Institute for the Study

of Labor (IZA)

1

I. Introduction

The early months of a newborn’s life are considered to have a profound and persistent role

in formulating children’s cognitive skills.1 For example, early months largely determine the

formation of synapses that permit communication between neurons, and the human brain

reaches 50 percent of its mature weight by six months of age (see Woodhead 2006 for an

extensive review). Having been born in periods of labor market buoyancy may help children’s

chances of having sufficient nutrient intake and a better living environment. Parents that

can afford to purchase toys, musical instruments, and other educational materials may help

stimulate children’s cognitive ability. However, when the labor market is in a downturn,

parents may have decreasing income levels, increasing mental stress, a deteriorating health

condition, and a worsening home environment. This might deprive children from getting not

only adequate nutrition, but also parental attention and emotional support, thus creating

negative consequences for the child’s development. At the same time, higher unemployment,

which is associated with a decrease in labor supply, might yield more available time for

parents, traditionally mothers, to engage in home production and take better care of the

child. Given the importance of early life in laying the foundation for children’s human

capital accumulation and long-term prospects,2 this paper investigates the link between

labor market conditions at birth and cognitive achievement during childhood.

The crucial empirical problem of analyzing the effect of parents’ labor market condi-

tions at birth on outcomes later in life is that unobserved heterogeneity could jointly affect

parental labor-market activity and children’s educational attainment. Those parents who

are more likely to be unemployed or lose jobs might have socioeconomic and home environ-

ments that are detrimental to children’s progress. One potential approach to address this

1E.g., UNICEF 2014, Woodhead 2006, Conti and Heckman 2012.2Epidemiologic studies provide further insight into why conditions during gestation affect adult health

outcomes. Heijmans et al. 2008 show that periconceptional exposure to famine creates persistent epigeneticdifferences that can last up to six decades after birth. Likewise, Denney and Quinn (2018) and Ravlic et al.(2018) find that adverse maternal factors at the time of fetal growth increase a child’s risk of developingchronic diseases in adulthood, including diabetes and obesity.

2

concern is to use a measure of labor market outcomes constructed at the macro level, such as

the unemployment rate (van den Berg, Lindeboom and Portrait 2006). However, the main

concern with the unemployment rate is that it is affected by both labor demand and supply,

and differences in parents’ labor supply behavior is related to differences in unemployment

rates and in children’s educational outcomes. Welfare programs in the U.S. that are normally

expanded during periods of high unemployment are criticized for distorting work incentives

and labor supply, resulting in a rise in the unemployment rate. For instance, Barro (2010)

and Hagedorn, Manovskii and Mitman (2015) argue that the unemployment rate after the

great recession of 2008-09 was artificially high because individuals were not returning to

work due to a generous expansion of unemployment benefits. Likewise, the unemployment

rate differs across race, ethnicity, and educational levels. The concentration of a particular

racial or an (un)educated group in a particular state affects the state’s unemployment rate.

Differences in children’s cognitive achievement associated with the unemployment rate could

be the result of the composition of the unemployed.

To address potential confounders, this paper constructs predicted labor demand

growth, which is arguably exogenous at the individual level. To do so, we calculate the

share of each industry in a state’s employment in the base year and interact it with the

growth rate of the corresponding industry at the national level. We construct separate pre-

dicted growth rates for men and women to examine if improved employment opportunities

for men and women have differential effects. Our empirical strategy is based on Bartik’s

(1991) “shift-share” approach,3 which has been applied by Katz and Murphy (1992), Page,

Schaller and Simon (2017), Aizer (2010), and Allcott and Keniston (2017), among others,

in different contexts. Our analysis uses children from the National Longitudinal Survey

3Goldsmith-Pinkham, Sorkin and Swift (2018) provide a detailed review of the approach. Our identifyingassumption is that industry shares in the base period are orthogonal to unobserved factors that affectchildren’s cognitive development process decades later. This assumption is not subject to the criticism inGoldsmith-Pinkham, Sorkin and Swift (2018) as it is hard to argue that industry shares in the base year1970 are correlated with educational outcomes in the late 1980s and the 1990s. We also analyze sensitivityof our findings in a robustness check by controlling for industry shares.

3

of Youth 1979 (NLSY79) Child and Young Adult that follows biological mothers in the

NLSY79. The survey is conducted every other year (from 1986 to 2014) and has a rich set

of information including widely used measures of cognitive ability such as math and reading

scores and family background. We use children born between 1978 and 2009, who were 5 to

14 years of age, a group whose math and reading scores are available in our data.

We begin by assessing the way predicted gender-specific employment growths interact

with parents’ labor market outcomes in the year of birth. We find that predicted male-

employment growth has a positive and significant effect on fathers’ labor supply and earnings

in the year of child birth. Furthermore, our estimates show that a one-percentage point rise

in male-specific labor demand growth reduces the likelihood of the family being in poverty by

around 1.4 percentage points. Our results show that predicted female-specific employment

growth positively and significantly affects the number of weeks mothers worked in the year of

child birth. However, the effect for mothers’ wages is not statistically different from zero. It

is worth noting that about a third of the women reported having zero weeks of employment

in the year of child birth in our sample, suggesting that those mothers may be more focused

on birth and other child related issues. In another set of results, we find that predicted

employment growths are highly correlated with unemployment rates.

We next estimate the effects of gender-specific employment growths on children’s test

scores. We find that a one-percentage point rise in male-specific labor demand growth leads

to a 0.018-0.033 standard deviation rise in math scores and a 0.025-0.032 standard devi-

ation increase in reading scores. Our estimates show the positive effect of female-specific

predicted labor demand growth, with the estimates showing increases of around 0.015 and

0.019 standard deviations in math and reading scores, respectively. That being said, we

cannot statistically distinguish these estimates from zero. The differential effects across gen-

der highlight the roles of both income and substitutability of parental care, as predicted

by Becker and Tomes (1986). This is also in line with the existing theoretical ambiguity

4

about whether a mother’s employment should create positive or negative effects on chil-

dren’s cognitive ability. To the extent that mother’s employment reduces the quality of care,

children’s cognitive development might be adversely affected. On the other hand, mother’s

employment and subsequent increase in family income lead to a positive effect on children’s

cognitive ability (e.g, Dahl and Lochner 2012). Rege, Telle and Votruba (2011), who inves-

tigate the micro effect of parental job loss that occurred when a child was in seventh grade

on short-term educational performance, provides support to this ambiguity. The authors

find an insignificant effect for mothers’ job loss on a child’s educational performance, but a

significant and negative effect for fathers’ job loss.

We assess the robustness of our findings to several specifications. We separately

estimate the effect of male-specific predicted employment growth on the child’s cognitive

achievement by the marital status of their mothers in the year of birth. We find stronger

effects on children of married mothers and insignificant effects on children of unmarried

mothers. If mothers were unmarried at the child’s birth, any improvement in men’s employ-

ment opportunities may not affect them. These findings could be viewed as a falsification

test. Even when we estimate the effect for male-specific employment growth, controlling

for female-specific employment growth to account for any other effect on child achievement

arising from interdependencies in spousal labor supply, the results are consistent. We also

separately estimate the effect by race. This subgroup analysis shows that effects are more

pronounced among white children. Moreover, as suggested by Jaeger, Ruist and Stuhler

(2018), we control for the lagged value of predicted male-specific growth to account for the

possibility of its serial correlation over time.

We extend our analysis to estimate the effect on the Peabody Picture Vocabulary

Test (PPVT), which measures receptive vocabulary.4 Our findings are similar to that of the

reading and math scores. We find a significantly positive effective of male-specific predicted

4Although the primary focus of this paper is to study the effect on a child’s cognitive development, wewere curious about any effects on children’s behavioral problems. We attempted to examine the effects onsuch problems, but found insignificant effects.

5

employment growth, but an insignificant effect of female-specific growth. In an attempt to

further understand the mechanism behind the positive effect of labor market growth at birth

in childhood, we investigate how labor market outcomes affect home environment in the

year of birth. The NLSY has information about the quality of home environment. We find

a significant and positive effect of male-specific predicted employment growth.

Finally, we explore a potential concern of whether couples self-select with regards

to choosing a time for childbearing. If the self-selection–as shown by Dehejia and Lleras-

Muney (2004)–occurs, our baseline model underestimates the effect of predicted employment

growths. Dehejia and Lleras-Muney (2004) argue that low-income parents’ skills deteriorate

relatively faster, and thus the opportunity cost of having a child for a low-income family

in labor market buoyancy is higher than that of a period of labor market slack. Likewise,

low-earning couples might find periods of labor market booms more preferable as they may

have enough financial resources to take care of the child. High-earning couples choose times

of labor market slack to raise a child as their opportunity cost of childbearing becomes lower.

We examine if predicted male-specific employment growth affects the behaviour of highly

educated and less educated married mothers differently when it comes to having a baby. We

do not find significant effects. Furthermore, we re-estimate our main findings using mother

fixed effects. Our objective is to compare differences in outcomes of siblings due to differences

in employment growth. Doing this allows us to net out time-invariant differences in family

background. Our main results are consistent.

In examining the link between labor market conditions at birth and outcomes later

in childhood, we contribute to three strands of the literature. First, this paper contributes

to a small body of literature that examines the effect of the business cycle during childhood

on later outcomes.5 For example, van den Berg, Lindeboom and Portrait (2006) study the

effect of the business cycle early in life on the individual mortality rate in adulthood. Using

5A number of research papers have studied the effects of shocks to pregnant mothers on child’s healthoutcomes (e.g., Black, Devereux and Salvanes 2016).

6

birth cohorts from 1908 to 1937 in the Netherlands, van den Berg et al. (2010) examine how

cognitive functioning in old age is affected by the business cycle in early life. Doblhammer,

van den Berg and Fritze (2013) employ a cross-national survey consisting of respondents

from 10 European countries to study the link between economic conditions at birth and

cognitive functioning among the elderly. Ruhm (2004) shows that mother’s employment

in the first three years of a child’s life leads to lower test scores at ages 5 to 6. Dehejia

and Lleras-Muney (2004) examine the association between the unemployment rate during

pregnancy and newborns’ health. However, there is little evidence on the effect of labor

market shocks in the year of a child’s birth on educational outcomes. Rao’s (2016) analysis

is centered around examining how average unemployment rates experienced in the years

between conception and age 15 are associated with outcomes in adulthood.

Second, our paper complements the existing literature that studies the role of job

displacement on children’s education (e.g, Hilger 2016, Stevens 1997, Pan and Ost 2014).

Much of the literature attempting to investigate the impact of labor market conditions fo-

cuses on individual layoffs, in an attempt to circumvent an endogeneity problem. These

findings represent the micro effect. We contribute to this literature by leveraging (an ar-

guably exogenous) employment growth to capture the general equilibrium or full effect of

unemployment.

Third, by examining the impact of female labor market conditions at birth, we con-

tribute to the literature that examines the link between mother’s employment and her child’s

educational outcomes. The literature does not reach a consensus with regards to the effect

of maternal participation in the workforce on children’s educational performance. James-

Burdumy (2005) finds that maternal employment in the year after birth has a limited effect

on a child’s test scores. Blau and Grossberg (1992) show that children whose mothers work

all weeks in the child’s first year of life have lower test scores. Baum (2003) finds negative

effects of maternal employment in the first year of a child’s life on cognitive development.

7

Likewise, others find no significant effect of maternal employment (e.g., Kalil and Ziol-Guest

2008). These differential findings in the literature could be the result of selection bias (see

Waldfogel 2002 and Bernal 2008), which we attempt to address.

The remainder of the paper unfolds as follows: Section II builds a conceputal frame-

work to show potential channels between labor market conditions at birth and educational

outcomes during childhood. Section III describes our data and Section IV our empirical

specification. We present results in Section V and apply a mother fixed effects model in

Section VI. Section VII concludes.

II. Conceptual Framework

We draw on Becker (1981) to illustrate how labor market conditions in the year of birth

impact the child’s cognitive production function, thus motivating the empirical exercise

below. In Becker’s model,6 a household maximizes its utility by consuming a bundle of goods

that includes a child as one component. The production of these goods requires inputs from

the household members in the form of time and effort as well as market-based goods. As

the value of household utility increases with a rise in the child’s quality, household members

focus on child care. However, the ability of household members or parents to improve the

well-being or the quality of the child is tied with family endowments and their labor market

opportunities. While total human capital up to a particular age is a dynamic, multistage,

and complex phenomena, a child’s early months, including the prenatal period, are crucial

in a child’s brain development (Woodhead 2006). The production of synapses that permits

communication between neurons peaks in the first year of life (Tierney and Nelson 2009).

To show how the formation of the child’s cognitive development is potentially shaped

by early life circumstances associated with parents’ labor market outcomes, we explain the

6Given the sample time period, this model is analogously appropriate.

8

link between three major inputs (parental time, market-based goods, and child care) of the

cognitive achievement production function and parental labor market outcomes in the year

of birth. To elucidate the role of parental time in a child’s achievement, consider an example

that the achievement of a 10-year-old child in the current year is influenced by time spent by

parents on various activities with the child since birth, which include helping with homework,

going on outings, and other physical and mental exercises. Likewise, note that market-based

goods purchased over time include health expenditures, housing, and educational resources.

Given the importance of these three inputs in child development, parents face com-

peting decisions: the tradeoff between engaging in home production and child care versus

participating in the labor market to increase the affordability of marked-based goods. Addi-

tionally, the tradeoff gives insight into how fathers’ and mothers’ job market opportunities

differ in fostering the child’s development, as women have traditionally played the leading

role when it comes to raising children. The psychology literature postulates that the major

channel through which job loss affects children’s cognitive development is through mental

stress and breakdown in emotional connection between the father and child (Elder, Nguyen

and Caspi 1985 and Christoffersen 2000). Better labor market conditions for fathers might

help strengthen that emotional connection. With the rise in father’s income associated with

employment growth, the family could provide nourishment and better living conditions to the

child. Similarly, an improvement in male job market conditions might help mothers reduce

their labor market production and increase their home production (Bertrand, Kamenica and

Pan 2015).

There are two major pathways through which improvement in maternal employment

opportunities affect child development. Employment growth creates income and substitution

effects with regard to the consumption of child care (Bernal and Keane 2011 and Ozabaci,

Henderson and Su 2014), assuming that the child is a normal good. A rise in own-wage

increases the relative price of home production and investment in the child. The decline in

9

homemade goods, such as meals and child care could be detrimental to the child’s psychology

and in the formation of cognitive ability. However, higher income enables better diets, better

access to health care, less stress, and goods that could stimulate a child’s brain. If the

reduction in home-goods is fully compensated by market goods, female-specific employment

growth may have no effect, as positive and negative effects offset each other.

In summary, we argue that labor market outcomes in the year of birth have profound

and persistent effects throughout childhood. Male-specific employment growth leads to a

positive impact on the formation of a child’s ability. On the other hand, since traditionally

mothers play a key role in raising children, the effect of female-specific employment growth

has an ambiguous effect. In the next section, we outline our empirical specification to esti-

mate the effects of predicted employment growths on child achievement, leveraging plausibly

exogenous variation in the local labor market.

III. Data

We use data from the National Longitudinal Survey of Youth (NLSY79) and the NLSY79

Child and Young Adult. The NLSY79 is a nationally representative longitudinal survey

of 12,686 young individuals (6,403 males and 6,283 females) born between 1957 and 1964.

In the first interview year 1979, respondents were 14-22 years old. The survey interviewed

individuals every year from 1979 through 1994 and every other year thereafter. The latest

available interview occurred in 2014. The data contain a rich set of information about

educational background, labor market outcomes, and ability as measured by the Armed

Forces Qualification Test (AFQT).

The NLSY79 Child and Young Adult cohort that was introduced in 1986 includes

biological children of women in the NLSY79. The survey is conducted every two years and is

ongoing. As of 2014, the latest survey, 11,521 children were born to women in the NLSY79.

10

Children appear in multiple surveys. Over 10,000 children appear in at least one survey. We

can link these children to their mothers in the NLSY79. These datasets are suitable for the

purpose of our study to examine the effect of local demand shocks at birth on educational

outcomes later in life. The data allow us to trace the birth state of the child and include

widely used measures of children’s cognitive achievements.

We present a histogram with an overlaid kernel density plot of birth cohorts in Figure

1.7 Most births occurred in the 1980s and 1990s. As our goal is to investigate the effect of

labor market shocks at birth, we need to identify the birth state of children. As mothers

have been interviewed since 1979, we could pinpoint the birthplace of only those children

who were born in or after 1979. Hence, we exclude children who were born before 1979,

which is about six percent of the total sample.

We use the restricted version of the data, as the publicly available version does not

have geocode information. The NLSY79 includes a nationally representative cross-sectional

sample along with an over-sample of Hispanic, black, economically disadvantaged whites,

and members of the military. In our analysis, we exclude the over-sampled children, basing

our analysis only on the cross-sectional sample. Furthermore, we use survey weights to make

our sample representative of the population.

Outcome Variable: Math and Reading Scores

We use the Peabody Individual Achievement Test (PIAT) that assesses academic achieve-

ment of children. It has been widely used as a measure of cognitive ability. The test is

administered every other year beginning with 1986. It is an age-appropriate test with an

increase in difficulty level from preschool to high school. The tests are assessed to children 5

to 14 years of age. The PIAT includes tests in math and reading. The PIAT Math assesses

a child’s mathematics achievement. The PIAT Reading Recognition measures a child’s word

recognition and pronunciation ability. Both tests contain 84 items. We normalized these

7The kernel density estimate is obtained via the Wang and van Ryzin (1981) kernel function for ordereddiscrete data with a plug-in bandwidth (see Chu, Henderson and Parmeter 2015).

11

scores (mean zero, standard deviation one) for our analysis.

Outcome Variable: Peabody Picture Vocabulary Test

The Peabody Picture Vocabulary Test (PPVT) “measures an individual’s receptive (hear-

ing) vocabulary for Standard American English and provides, at the same time, a quick

estimate of verbal ability or scholastic aptitude” (Dunn and Dunn 1981). The test has been

designed targeting children as young as two and a half years to 40-year-old adults. In the

NLSY79 Child and Young Adult survey, children 4-5 and 10-11 years of age took the PPVT.

Additionally, the test was administered to some children falling in other age groups. The

test includes 175 stimulus words along with 4 black-and-white drawings for each stimulus

word for which the goal is to choose the drawing that best represents the meaning of the

stimulus word. In our analysis, we normalize the total score.

Outcome Variable: Home Environment

The NLSY79 Child and Young Adult survey includes the Home Observation Measurement

of the Environment-Short Form (HOME-SF) that is intended to assess the child’s home

environment. The HOME-SF includes a number of assessment items, which respondents

are asked to report. These items are considered to be an input in child development. For

example, the questionnaire items include indicators for whether the “Child gets out of house

4 times a week or more,” “Child has 3 children’s books,” and “Child has one or more cuddly,

soft or role-playing toys.” Based on responses to these questions, the total raw score for the

HOME-SF is calculated. In addition to the total raw score, the NLSY79 Child and Young

Adult includes cognitive stimulation and emotional support scores based on these assessment

items. This paper uses all three measures - total raw score, stimulation support score, and

emotional support score. We normalize these scores and restrict our analysis to children aged

0 to 12 months, as our aim is to assess how labor market outcomes impact the children’s

home environment during the year of birth.

After matching children to mothers, we construct variables related to mothers. From

12

1979 to 1994, mothers’ labor market outcomes and state of residence are available annually.

Thereafter, they are available every other year. We use the average of the subsequent year

and preceding year for each missing year to calculate income and other labor market variables.

Likewise, to identify the state in each missing year, we use information from the subsequent

year. For example, when a mother is not interviewed in 1995, we use her location in 1996

for the year 1995. In cases where the geocode information is missing in the next year, we

use it from the previous year. Likewise, if mother’s education information is missing in some

surveys, we construct the education variable as the highest education achieved up to the

point of the survey, using the current and previous surveys.

We merge these data with the unemployment rate data extracted from the Bureau of

Labor Statistics. Likewise, we use the March Current Population Survey (CPS) each year

from 1977 to 2010 and the 1970 U.S. decennial census to create the predicted employment

growth rate as described in the next section. Descriptive analysis is provided in Table 1.

IV. Empirical Strategy

Our empirical analysis exploits variation in labor market shocks at birth over space and time

and individual data to measure effects on children’s educational outcomes. We pool data in

a cross-sectional format. Our unit of analysis is a child-year, that is, we have observations for

the same child in different years.8 As parents’ individual labor market outcomes in the year

of birth could be correlated with children’s cognitive development later in life, our approach

here is to find a measure of labor market outcome that is exogenous at the individual level.

For example, children, whose parents are unemployed, underemployed or out of the labor

force, may have other adverse family and socioeconomic backgrounds that could negatively

affect their educational achievements. To avoid our analysis being diluted by unobserved

8We also consider a child as a unit of analysis in robustness checks in Appendix (Table B1). We collapsetest scores at the child level. In an alternative specification, we also limit our analysis to a child’s first testscore.

13

individual heterogeneity, we focus on macroeconomic variables. One approach to estimate

the effect of labor market conditions at birth on educational outcomes later in life is to use

the following equation:

Ai = α + β1Ust + λXi + ρZi + γt + ςb + ψs +Bl + εi, (1)

where i indexes a child-year and A represents a math or reading score. Although a child can

appear repeatedly in our data, the variable of interest Ust, the state-level unemployment rate

in the year of a child’s birth t, does not vary by child. We use our data in a cross-sectional

format. X includes a vector of children’s characteristics which includes age, age-squared, and

race. Z is a vector of mother characteristics which capture family backgrounds affecting child

development. It includes mother’s age, education, and AFQT scores9 (the latter administered

in 1980). We control for the number of siblings, as having many siblings divides parents’

resources.10 These variables are measured in each survey year. We include the mother’s

average permanent income, which is calculated as the average of mother’s reported real

wages11 each year up to the point of the survey year.12 γt are year-fixed effects, which are

intended to capture shocks that are common to all states. For example, this could be any

expansion or contraction in a federal welfare program or federal educational policy that

equally affects all states. ςb is a vector of state-of-birth fixed effects. This controls for state-

specific differences such as education and health care during the year of birth. ψs represents

a vector of state-of-residence fixed effects, which account for the time-invariant state-specific

differences. Buckles and Hungerman (2013) show that season of birth is associated with

outcomes in adulthood; to control for potential seasonality, we use the month-of-birth fixed

9We rescale the scores (dividing by 10,000) for simplicity of interpretation.10It could be argued that sibling size is affected by prior labor market conditions and is endogenous. We

re-estimate our model excluding it and our main results do not change.11We use the Consumer Price Index for All Urban Consumers (CPI-U) to convert wages into real values.

We use 1990 as the base year. We rescale wages (dividing by 10,000).12We use this measure instead of mother’s wage during the year of birth as wages in the year of birth could

be noisy due to a mother’s irregular participation in the labor market due to issues related to childbirth.Further, when we re-estimate our model excluding it, the main results do not change.

14

effects Bl. The error term (εi) contains unobserved factors that vary by child-year. α, β1,

λ, ρ, γ, ς, ψ and B are parameters to be estimated, where β1 is the parameter of interest.

We cluster standard errors at the state-of-birth level. We use survey weights to make our

sample representative of the population.

Our measure of the unemployment rate at the state level could be problematic as it

is affected by both labor demand and supply. For example, the unemployment rate at a

particular time could be high if there is an increase in labor force participation. Another

drawback of the unemployment rate is that it does not include discouraged workers, and those

who are not participating in the labor market. As a result, it is correlated with individual

characteristics that could bias estimates.

To create a measure of labor market conditions that is arguably exogenous at the

individual level, we use the predicted employment growth rate which will replace the unem-

ployment rate in Equation 2, following Bartik’s (1991) “shift-share” approach.13 This growth

rate basically captures the exogenous (to the individual) changes in state labor demand. We

classify total employment into 17 categories in the base year and calculate the share of each

industry at the state level in the base period (1970). We also create the annual growth rate

of each industry at the national level. We multiply the growth rate of each industry in a

particular year with the corresponding share in the base period, and sum over industries to

create the predicted employment growth for that particular year. Specifically, we define the

employment growth rate (Lst) as:

Lst =17∑k=1

$ks,1970∆N

kt , (2)

where $ks,1970 is the share of employment in industry k14 in state s in the base year and

13Schaller (2016) uses a similar approach to study the effect of the unemployment rate on the birth rate.14We follow Schaller (2014) and Katz and Murphy (1992) to create 17 industry categories. We use

the variable “IND1950” that uses the 1950 Census Bureau industrial classification scheme to consistentlyclassify industries. The industry categories are: (1) Agriculture, Forestry and Fishing; (2) Mining; (3)Construction; (4) Low Tech Manufacturing; (5) Basic Manufacturing; (6) High Tech Manufacturing; (7)

15

∆Nkt is the growth rate of national employment in industry k in year t.15 In other words,

we weight the share of employment in each industry in each state by the growth rate of

the corresponding industry at the national level, and sum over industries. We use the 1970

Census data to create the industry share in the base year.16 In a robustness check, we also

use the 1980 Census. To calculate the annual growth rate of the industry, we use the March

Current Population Survey (CPS). For the March CPS, we use the survey weight to create

the employment level for the total population for each year.

We also construct gender-specific employment growth rates, as male and female job

market opportunities could affect children’s educational outcomes differently. For example,

the literature that attempts to understand the effect of parental job loss on children’s educa-

tional outcomes mainly focuses on father’s job loss. Father’s job loss is expected to negatively

affect children’s educational outcomes. The effect of mother’s job loss or unemployment is

ambiguous. In the spirit of this literature, we can expect to have a positive effect of predicted

male employment growth, while female’s employment growth can have an ambiguous effect.

To construct gender-specific employment growth rates (Lgst), we use the following approach:

Lgst =17∑k=1

$kgs,1970∆N

kt (3)

where g ∈ {male, female} and $kgs,1970 is the ratio of each gender employment in an industry

to the total employment of each gender in a state in the base year.

Using this data, we can modify Equation 1 by replacing the unemployment rate with

Transportation; (8) Telecommunications; (9) Utilities and Sanitary Services; (10) Wholesale Trade; (11)Retail Trade; (12) Finance, Insurance, and Real Estate; (13) Business and Repair Services; (14) PersonalServices; (15) Entertainment and Recreational Services; (16) Professional and Related Services; and (17)Public Administration.

15We prefer the growth rate to the level of employment as the former is better able to capture labor marketshocks. Additionally, growth rates are not sensitive to the size of states and are comparable across states.

16We prefer the 1970 Census to 1980 because variation in the “shift-share” approach is mainly driven bystate industry shares in the base year (see Goldsmith-Pinkham, Sorkin and Swift 2018). Industry sharesobserved in 1970, a decade or earlier than the births of children in our sample, further mitigate any con-cern about the potential correlation between predicted employment growths and confounders of children’seducational performance.

16

the predicted employment growths we created. This leads to the following model:

Ai = α + β1Lst + λXi + ρZi + γt + ςb + ψs +Bl + εi, (4)

where we can also replace Lst with Lgst and estimate the effect of employment or gender-

specific employment growth in the year of a child’s birth t. β1 is the parameter of interest

and it measures the effect of predicted employment growth (or gender-specific predicted

employment growth) in the year of birth on children’s math or reading scores. We cluster

standard errors at the state-of-birth level. Clustering at both the state-of-birth and child

level yields standard errors close to those of our baseline estimates. The main identifying

assumption is that other unobserved factors that affect children’s educational outcomes are

not correlated with predicted employment growths in the year of birth. It is important to

note that the major variation in our predicted growth rates stem from the base-year industry

share. Hence, it is unlikely for contemporaneous business cycle or family decisions to bias

our results. Further, it is worth noting that Dehejia and Lleras-Muney (2004) argue that

high income families choose periods of high unemployment and low income families periods

of lower unemployment to have a child. If such sorting takes place, our baseline model

underestimates the effect, implying that the positive effect we find here is not attributable

to the self-section. Nonetheless, to mitigate this potential concern that families may have

self-selected a time for childbearing, we employ a mother fixed effects model in Section VI.

V. Results

A. First Stage Estimates

Before we estimate our main specification, we examine relationships between predicted em-

ployment growths and parents’ labor market outcomes. We limit our analysis to parents

17

of children whose standardized test scores are observable in our sample. The results are

reported in Table A1. We find positive and significant effects for predicted male-specific em-

ployment growth on the number of weeks worked by fathers and fathers’ wages in the year

of the child’s birth. One caveat of this analysis is that in our sample, fathers’ labor market

outcomes are reported by mothers and the sample has several missing values for these vari-

ables, especially earnings. We also look at the effect of predicted female-specific employment

growth on mothers’ labor supply and earnings in the year of the child’s birth. Our estimates

show a significant and positive effect on the number of weeks worked by mothers.17 The ef-

fect on wage is insignificant.18 It is worth noting that more than a third of the women report

having zero weeks of employment in the calendar year of the child’s birth. Additionally, we

investigate whether gender-specific employment growths could affect family poverty status

(Table A2). We find that an improvement in men’s labor market opportunities reduces the

likelihood of the family being in poverty in the year of the child’s birth. However, we do not

find a significant effect on poverty for female-specific employment growth. Additionally, we

examine correlations between the unemployment rate and predicted employment growths.

We present graphical evidence in Figure A1. Likewise, we regress the unemployment rate

on predicted employment growths, separately, controlling for state and year fixed effects. As

reported in Table A3, we find a statistically significant association between these.

As previously mentioned, we focus on reduced form estimates below for three reasons.

First, for those children born after 1994, the data do not have parents’ labor market outcomes

in alternate years, as the survey is conducted biannually. Second, as the mother reports her

spouse’s labor market outcomes, they could be prone to measurement error. Third, predicted

employment growth operates through multiple inputs to affect the achievement production

function. As Todd and Wolpin (2003, 2007) and Cunha and Heckman (2007) discuss in

17The estimate is insignificant when we focus on the intensive margin of labor supply by conditioning ourestimate on those having positive weeks of work.

18We also examine the effect of gender-specific employment growth on these labor market outcomes,controlling for a respective spouse’s employment growth. The results are qualitatively similar.

18

detail, given the role of multiple inputs in child achievement, using instrumental variables is

not an appropriate approach.

B. Main Estimates

We begin by estimating Equation 1 to examine the effect of the unemployment rate on chil-

dren’s math and reading scores. The results are reported in Table 2. The unemployment

rate has a small positive effect on both math and reading scores. A one-percentage point

rise in the state unemployment rate leads to a 0.002 standard deviation increase in math

scores and a 0.009 standard deviation rise in reading scores. However, they are not statisti-

cally different from zero.19 Though these estimates provide initial evidence of labor market

conditions at birth on a child’s test scores, this evidence is inconclusive. It is not surprising

to have these findings, given that in theory, the effect of the unemployment rate in the year

of birth on children’s test scores is unclear. The unemployment rate is sensitive to both the

composition of labor force and (non)participation in the labor force,20 which could be corre-

lated with other state- and individual-level unobserved factors that affect child development,

both positively and negatively. For example, in a particular state, the number of individuals

from certain groups (e.g., less-skilled or less educated) could change faster than the average

in that state. It has been shown that for the less skilled/educated, the unemployment rate

is higher than that of the overall unemployment rate. Hence, their disproportionate growth

(decline) affects the state’s unemployment rate differently. Further, their children are ex-

pected to have worse educational outcomes as compared to the average child. Therefore,

any changes to a child’s test scores associated with the unemployment rate could partly be

ascribed to other unobserved family backgrounds rather than being the true effect of labor

market conditions. Such a composition change may result in a negative association between

19Among control variables, we find a positive effect of age on math scores and negative effect on readingscores. We find it puzzling why we would have opposite effects of age on these test scores.

20 Murphy and Topel (1997) using the Current Population Survey (CPS) data from 1968 to 1995, explainwhy the unemployment rate increasingly became a poor measure of labor market conditions.

19

the unemployment rate and a child’s test scores.

Besides being prone to the individuals’ withdrawal from the labor force, the unem-

ployment rate is affected by participation in the labor force. For example, in the 1980s and

1990s, more women participated in the labor force on the back of decreasing discrimination

against women in the labor market, which might have pushed the unemployment rate up.

If that was the case, the unemployment rate might have increased even if the actual labor

market conditions were improving. In that case, we should expect the positive relationship

between the unemployment rate and a child’s test scores.

As widely shown in the literature, the generosity of unemployment insurance (UI) ben-

efits influences the unemployment rate, promoting individuals to stay unemployed longer.

UI benefits are also shown to be helpful in children’s education (Regmi 2019). In line with

these strands of the literature, it could be argued that UI benefits are being positively cor-

related both with the unemployment rate and a child’s test scores. Likewise, the aggregate

unemployment rate could mask the differential effects of a father’s and a mother’s unem-

ployment on child development. For example, Page, Schaller and Simon (2017) find that an

improvement in a mother’s employment opportunities lead to a decline in a child’s health.21

If these omitted factors were working in different directions as in theorey, we would expect

the direction of bias in the OLS estimates to be ambiguous.

To address the limitation of the unemployment rate, we exploit labor demand shocks,

as captured by the predicted employment growth rate described above, instead of the unem-

ployment rate. Our estimates could be viewed as reduced form coefficients. We begin our

analysis using the overall employment growth rate. As reported in Table 3, an improvement

in overall labor market opportunities positively affects children’s educational outcomes. A

one-percentage point growth in the predicted employment rate during the year of birth in-

21Likewise, Lindo, Schaller and Hansen (2018), who study the effect of labor market conditions on childmaltreatment using county-level data from California, show that the female-specific employment growthleads to a statistically significant increase in child maltreatment reports.

20

creases math scores of children aged 5 to 14 years by around 0.019 standard deviations, and

reading scores by around 0.026 standard deviations.

It is well established in the literature that men and women’s labor market conditions

affect children’s educational outcomes differently. For example, Rege, Telle, and Votruba

(2011), using Norwegian register data, show that it is father’s mental stress that is detri-

mental to children’s educational performance. Table 4 shows our results. We find that better

labor market opportunities for males significantly improve children’s cognitive development.

Furthermore, we find that female-specific labor market growth has positive effects on both

math and reading scores. The magnitudes of these estimates are close to male-specific es-

timates. However, we cannot statistically distinguish these estimates from zero.22 Weaker

first-order effects of the predicted female-specific employment growth on labor market out-

comes may have resulted in its statistically weaker effects on test scores.

In our analysis above, we examine the effect of male-specific employment growth on all

children. However, for those children born to unmarried mothers, we would not expect child

outcomes to be affected by improvement in men’s labor market opportunities. Therefore, we

also separated children by mother’s marital status in the year of birth. We separately esti-

mate the effect of male employment growth on children of married and unmarried mothers.

Panel A of Table 5 contains the results. We do not find a statistically significant effect on the

children of unmarried mothers. However, we find stronger effects for the children of married

mothers. For comparison, we also examine effects of female-specific employment growth for

the children of married mothers and unmarried mothers, separately. The estimates are sta-

tistically insignificant for both the groups (Panel B of Table 5). Our analysis on unmarried

mothers could serve as a possible falsification check. If other unobserved factors were driving

our results, we should expect to get a significant effect of the predicted male employment

growth rate on children of unmarried mothers.23 In summary, these results help corroborate

22We also estimate our model without controlling for mother characteristics. The results reported in TableB2 in Appendix B are qualitatively similar.

23As noted above, we focus on the cross-sectional sample for our analysis. In a robustness check, we

21

our identifying assumption that labor market shocks calculated in this paper are exogenous.

For the rest of our analysis, we use children of mothers who were married during the year

of birth of the child when we estimate the effect of the male-specific predicted employment

growth rate. While investigating the effect of the female-specific employment growth rate,

we continue to use the full sample.

C. Alternative Outcome

In this subsection, we examine the effect on an alternative outcome variable. We use the

Peabody Picture Vocabulary Test (PPVT), which is typically administered to children ages

4 to 5 years and 10 to 11 years, through there are a few children from other age groups

who took this test. We use all children under 15 years of age who took the PPVT. We

separately estimate the gender-specific employment growth rates on the child’s PPVT score.

Table 6 contains the results. We find that improvement in labor market opportunities for

men positively affects test scores. Our estimates show that a percentage point rise in the

male-specific predicted employment growth rate in the year of birth increases the child’s

PPVT score by around 0.023 standard deviations. However, we find a comparatively lower

magnitude when we use the female-specific employment growth rate, and the effect is not

statistically different from zero. These findings strengthen our earlier estimates.

D. Home Environment

In order to understand the mechanisms through which improved/worsening labor market

conditions in the birth year operate to shape children’s cognitive development, we examine

the effect on home environment in the year of birth. The NLSY79 Child and Young Adult

contains three measures of home environment based on the Home Observation Measurement

estimate the effects for the full sample. We find the point estimates and standard errors similar to those ofour baseline estimates.

22

of the Environment (HOME)-Short Form: (i) total HOME score, (ii) cognitive stimulation

score, and (iii) emotional support score. These scores are calculated on the basis of responses

from parents to questions related to the nature and quality of the child’s home environment,

which are considered to be crucial in a child’s development. For example, it measures the

availability of materials for learning and maternal acceptance of and involvement with her

child, among others. We estimate the effect of the male-specific employment growth rate

on all three measures. We report the results in Table 7. We find that increasing labor

market opportunities for men lead to a significant improvement in overall home environment

and emotional support. The psychology literature postulates that father’s job loss creates

a breakdown in the emotional relationship between father and child, adversely affecting the

child’s education (Elder, Nguyen and Caspi 1985 and Christoffersen 2000). In line with

this literature, our results suggest that better emotional support to a child in the year of

birth, stemming from an improvement in men’s employment opportunities, positively affects

a child’s cognitive development.

E. Heterogeneity

Age. In this subsection, we examine if the effect of labor market shocks at birth fade with

age. To understand the persistence of the effect, we divide data between two age groups: 5

to 10 years and 11 to 14 years. Table 8 contains the results. We find statistically significant

effects on both math and reading sores for both groups of children.

Race. As seen in the literature, human capital formation patterns often differ by ethnicity

and race. Therefore, we examine the effect by race: white, black, and Hispanic. As reported

in Table 9, the effect is concentrated among white children. The coefficients are larger

and standard errors are smaller for this group. For black and Hispanic children, we find

insignificant effects. We also create race-specific employment growth rates for men. When

we use these growth rates, our results are similar (Table B5).

23

F. Robustness Checks

In this subsection, we carry out additional robustness checks. First, we perform an analysis

to examine whether pre-existing trends are driving our results. As in common in the empir-

ical literature of event studies, we include several lags and leads of predicted male-specific

employment growth in our main specification. In particular, we examine its effects at ages

-3, -2, -1, 0, 1, 2, and 3. We present the coefficients in Figure A2 with 95 percent confidence

intervals. We have insignificant effects prior to the year of birth and significant effects during

the year of birth. The effects decrease after the first year.

Second, we examine whether or not subsequent labor market conditions could offset

the effects experienced at birth. To do so, we re-estimate Equation 4 by including the average

of the predicted male employment growth rates experienced by the child from the second

year to the year of each survey (test-taking time). For example, if a child is surveyed at age

eight, we calculate the average from the time s/he is two to seven. The results, reported in

(Table 10), are similar to the baseline estimates.

Third, we estimate the effect of both male- and female-specific employment growths

by including both types of employment growths in the same regression. This is intended to

account for any other effect on child achievement arising from interdependencies in spousal

labor supply. As reported in Table A4, using both gender-specific employment growths in our

main specification produces point estimates of around 0.048 and 0.046 standard deviations

for male-specific employment growth for math and reading, respectively. In comparison, the

estimates in our main specification were around 0.033 and 0.032 standard deviations (Table

5) for math and reading, respectively. For female-specific employment growth, we continue

to find statistically insignificant effects, but the signs of the estimates flip.

Fourth, as shown in Figure 1, most of the births in the sample occurred in the 1980s

and 1990s. According to the Center for Human Resource Research (2002), most of the women

had passed their primary childbearing age by 2000. The youngest woman in 2000 was 36

24

years old. It could be argued that births after 2001 occurred as a result of female childbearing

postponement (or unintentionally) and children born after 2001 might be different from the

average sample. To address this concern, we estimate Equation 4 by restricting the sample

to children born before 2001 (Table A5). The results are consistent.

When we create the national growth rate of employment in an industry, we use the

CPS March sample. Our time-series variation in the industry’s growth at the national level

comes from the employment observed in March. One possible concern is that those children

who were born towards the end of the year could be less influenced by labor market conditions

observed in March. To deal with this concern, we re-estimate our baseline model excluding

children born after June. As presented in Table A6, we find stronger effects. Furthermore,

we re-estimate our main findings using 1980 as the base year, applying the decennial Census

1980 data. We find similar estimates (available upon request).

Additionally, we re-estimate our main model by controlling for the lagged value of

predicted male employment growth to control for serial correlation and other dynamics

related to the labor market adjustment process (see Jaeger, Ruist and Stuhler 2018). We

also control for state-level industry shares in the base year24 in our main specification to

address the possibility that our results are affected by the differential industry shares across

states.

Finally, we re-examine the effect of the unemployment rate during the year of birth on

children’s cognitive development, instrumenting the unemployment rate by predicted male-

specific employment growth. The results are presented in the Appendix (Table A8). Our

estimates show that a higher unemployment rate negatively affects children’s educational

outcomes.

24When we use state-level industry shares, we must exclude birth year fixed effects. The results, whichare reported in Table A7, are consistent.

25

VI. Self-Selection and Mother Fixed Effects

A potential concern with our identification strategy could be that parents, depending on

their income, might react differently to local labor market conditions when it comes to their

plan to have a child. There is a possibility that wealthy parents face higher opportunity

costs of having a child during improved labor market conditions. As a result, they may

decide to raise a child when the labor market faces slack. On the other hand, low-income

or financially constrained families may prefer the timing of improvement in labor market

conditions to raise a child, as their ability to provide necessary resources to the child im-

proves. This is especially true when financial markets are not perfectly efficient and couples

face borrowing constraints. Hence, we analyze whether highly and less educated mothers25

respond differently to predicted labor market growths. We estimate the following model:

Birthjt = α + β1Lg,st ∗ Collegejt + β2Lg,st + β3Collegejt + λXjt + γt + ψs + εjt, (5)

where Birthjt is an indicator variable if married mother j has a child under one year of age in

calendar year t. This measures whether a mother gave birth to a child in the calendar year.

Lg,st is the male-specific predicted employment growth rate in state s in the calendar year t

and Collegejt is a dummy variable that takes the value of one if a mother has a bachelor’s

degree or beyond. As reported in Table A9, we do not find evidence of self-selection into child

bearing by predicted male-employment growth. We also estimate the differential effect of

predicted female-specific employment growth for married mothers. We find an insignificant

effect, suggesting that an improvement in female’s labor market opportunities does not

affect child bearing decision differently among highly and less educated married mothers.

However, when we estimate the effect for single mothers, we find that higher educated

mothers are less likely to give birth during improved labor market conditions compared

to less educated mothers. Overall, our estimates suggest that our main findings–predicted

25We define highly educated mothers as those having a bachelor’s degree or beyond.

26

male-specific employment growth positively affects children’s cognitive development–are not

driven by the sorting of women into giving birth by educational attainment.

In a further robustness check to examine if such sorting takes place and our results are

potentially biased, we re-estimate our main model using mother-fixed effects. Specifically,

we use the following model to examine the effect of predicted male employment growth on

a child’s cognitive achievement

yij = α + β1Lmale,st + λXij + ρZij + φj + γt + ςb + ψs +Bl + εij, (6)

where i indexes child, j indexes mother, Lmale,st is the male-specific predicted employment

growth rate in state s in the year of birth t, φj is a vector of mother-fixed effects and the

remaining variables are defined as before. In this specification, we exploit variation within

siblings. This controls for time-invariant family backgrounds and genetics, including family

preferences with regards to timing of birth. As before, we restrict the sample to those

children whose mothers were married at the time of birth. We report the results in Table 11.

Our findings are qualitatively similar. This suggests that our baseline results are unlikely to

be biased by parents’ choosing the timing of child birth.

VII. Conclusion

We have presented evidence that local labor market conditions during the year of birth af-

fect cognitive ability in childhood. To address omitted heterogeneity that jointly determine

parents’ labor market outcomes and children’s development, we construct gender-specific

predicted employment growth rates, following Bartik’s (1991) “shift-share” approach. Specif-

ically, we construct this measure by interacting the cross-sectional variation in an industry’s

share in state-employment in the base year with the growth rate of that industry at the na-

tional level. Our estimation focuses on examining children’s math and reading scores when

27

they are 5 to 14 years of age.

We find that improvement in male employment opportunities in the year of the child’s

birth helps improve cognitive ability during childhood. However, we do not find significant

effects for women’s improved labor market opportunities on a child’s cognitive development.

Our findings suggest the competing role of the income effect and parental care. Our re-

sults are robust to different specifications and alternative outcomes. We find a positive and

significant effect for male-specific predicted employment growth on the Peabody Picture

Vocabulary Test (PPVT) scores, but an insignificant effect for female-specific predicted em-

ployment growth. We find similar results using a mother fixed effects model that nets out

time-invariant family backgrounds and other family characteristics.

Our findings carry important policy implications regarding the improvement of long-

term prospects of children. In line with the extensive literature that shows the importance of

early childhood education, one implication of this paper is that increasing social assistance

during times of high unemployment might help shield children from adverse labor market

outcomes. Likewise, the literature suggests that a father’s health, especially his mental

stress, is a major pathway through which his job displacement affects children’s educational

outcomes. Besides boosting transfer through social insurance programs such as unemploy-

ment insurance benefits, expanding health insurance coverage (both physical and mental)

to the unemployed during labor market downturns could help lessen the the effect of labor

market shocks on their children’s cognitive development.

28

References

Aizer, Anna. 2010. “The Gender Wage Gap and Domestic Violence.” American Economic

Review, 100(4): 1847–59.

Allcott, Hunt, and Daniel Keniston. 2017. “Dutch Disease or Agglomeration? The Local

Economic Effects of Natural Resource Booms in Modern America.” Review of Economic

Studies, forthcoming.

Barro, Robert. 2010. “The Folly of Subsidizing Unemployment.” The Wall Street Journal,

August, 30.

Bartik, Timothy J. 1991. Who Benefits from State and Local Economic Development

Policies? W.E. Upjohn Institute for Employment Research.

Baum, Charles L. 2003. “Does Early Maternal Employment Harm Child Development?

An Analysis of the Potential Benefits of Leave Taking.” Journal of Labor Economics,

21: 409–448.

Becker, Gary. 1981. A Treatise on the Family. National Bureau of Economic Research, Inc.

Becker, Gary S, and Nigel Tomes. 1986. “Human Capital and the Rise and Fall of

Families.” Journal of Labor Economics, 4(3): 1–39.

Bernal, Raquel. 2008. “The Effect of Maternal Employment and Child Care on Children’s

Cognitive Development.” International Economic Review, 49(4): 1173–1209.

Bernal, Raquel, and Michael P. Keane. 2011. “Child Care Choices and Children’s Cog-

nitive Achievement: The Case of Single Mothers.” Journal of Labor Economics, 29(3): 459–

512.

Bertrand, Marianne, Emir Kamenica, and Jessica Pan. 2015. “Gender Identity and

Relative Income within Households.” Quarterly Journal of Economics, 130(2): 571–614.

29

Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes. 2016. “Does Grief Trans-

fer across Generations? Bereavements during Pregnancy and Child Outcomes.” American

Economic Journal: Applied Economics, 8(1): 193–223.

Blau, Francine D, and Adam J Grossberg. 1992. “Maternal Labor Supply and Chil-

dren’s Cognitive Development.” Review of Economics and Statistics, 74(3): 474–481.

Buckles, Kasey S., and Daniel M. Hungerman. 2013. “Season of Birth and Later

Outcomes: Old Questions, New Answers.” Review of Economics and Statistics, 95(3): 711–

724.

Center for Human Resource Research. 2002. “NLSY79 Child and Young Adult.”

Columbus: The Ohio State University, CHRR NLS User Services.

Christoffersen, M. N. 2000. “Growing Up with Unemployment: A Study of Parental

Unemployment and Children’s Risk of Abuse and Neglect Based on National Longitudinal

1973 Birth Cohorts in Denmark.” Child Development, 7(4): 421–438.

Chu, C.-Y., Henderson D.J., and C.F. Parmeter. 2015. “Plug-in Bandwidth Selection

for Kernel Density Estimation with Discrete Data.” Econometrics, 3: 199–214.

Conti, Gabriella, and James Heckman. 2012. “The Economics of Child Well-Being.”

National Bureau of Economic Research, Inc NBER Working Papers 18466.

Cunha, Flavio, and James Heckman. 2007. “The Technology of Skill Formation.” Amer-

ican Economic Review, 97(2): 31–47.

Dahl, Gordon B., and Lance Lochner. 2012. “The Impact of Family Income on Child

Achievement: Evidence from the Earned Income Tax Credit.” American Economic Review,

102(5): 1927–56.

Dehejia, Rajeev, and Adriana Lleras-Muney. 2004. “Booms, Busts, and Babies’

Health.” Quarterly Journal of Economics, 119(3): 1091–1130.

30

Denney, Jeffrey M., and Kristen H. Quinn. 2018. “Gestational Diabetes: Underpinning

Principles, Surveillance, and Management.” Obstetrics and Gynecology Clinics of North

America, 45(2): 299 – 314.

Doblhammer, Gabriele, Gerard J. van den Berg, and Thomas Fritze. 2013. “Eco-

nomic Conditions at the Time of Birth and Cognitive Abilities Late in Life: Evidence from

Ten European Countries.” PLOS ONE, 8(9): 1–12.

Dunn, Lloyd M., and Douglas M. Dunn. 1981. “Peabody Picture Vocabulary

Test—Revised.” Circle Pines, MN: AGS.

Elder, Glen H., Jr., Tri van Nguyen, and Avshalom Caspi. 1985. “Linking Family

Hardship to Children’s Lives.” Child Development, 56(2): 361–375.

Goldsmith-Pinkham, Paul, Isaac Sorkin, and Henry Swift. 2018. “Bartik Instru-

ments: What, When, Why, and How.” National Bureau of Economic Research 24408.

Hagedorn, Marcus, Iourii Manovskii, and Kurt Mitman. 2015. “The Impact of

Unemployment Benefit Extensions on Employment: The 2014 Employment Miracle?”

National Bureau of Economic Research Working Paper 20884.

Heijmans, Bastiaan T., Elmar W. Tobi, Aryeh D. Stein, Hein Putter, Gerard J.

Blauw, Ezra S. Susser, P. Eline Slagboom, and L. H. Lumey. 2008. “Persistent

epigenetic differences associated with prenatal exposure to famine in humans.” Proceedings

of the National Academy of Sciences, 105(44): 17046–17049.

Hilger, Nathaniel G. 2016. “Parental Job Loss and Children’s Long-Term Outcomes: Ev-

idence from 7 Million Fathers’ Layoffs.” American Economic Journal: Applied Economics,

8(3): 247–83.

Jaeger, David A, Joakim Ruist, and Jan Stuhler. 2018. “Shift-Share Instruments and

the Impact of Immigration.” National Bureau of Economic Research 24285.

31

James-Burdumy, Susanne. 2005. “The Effect of Maternal Labor Force Participation on

Child Development.” Journal of Labor Economics, 23(1): 177–211.

Kalil, Ariel, and Kathleen M. Ziol-Guest. 2008. “Parental employment circumstances

and children’s academic progress.” Social Science Research, 37(2): 500 – 515.

Katz, Lawrence F., and Kevin M. Murphy. 1992. “Changes in Relative Wages, 1963-

1987: Supply and Demand Factors.” Quarterly Journal of Economics, 107(1): 35–78.

Lindo, Jason M., Jessamyn Schaller, and Benjamin Hansen. 2018. “Caution! Men

not at Aork: Gender-Specific Labor Market Conditions and Child Maltreatment.” Journal

of Public Economics, 163(C): 77 – 98.

Murphy, Kevin, and Robert Topel. 1997. “Unemployment and Nonemployment.” Amer-

ican Economic Review, 87(2): 295–300.

Ozabaci, Deniz, Daniel J. Henderson, and Liangjun Su. 2014. “Additive Nonpara-

metric Regression in the Presence of Endogenous Regressors.” Journal of Business &

Economic Statistics, 32(4): 555–575.

Page, Marianne, Jessamyn Schaller, and David Simon. 2017. “The Effects of Ag-

gregate and Gender-Specific Labor Demand Shocks on Child Health.” Journal of Human

Resources, forthcoming.

Pan, Weixiang, and Ben Ost. 2014. “The impact of parental layoff on higher education

investment.” Economics of Education Review, 42(C): 53–63.

Rao, Neel. 2016. “The Impact Of Macroeconomic Conditions In Childhood On Adult Labor

Market Outcomes.” Economic Inquiry, 54(3): 1425–1444.

Ravlic, Sanda, Nikolina Skrobot Vidacek, Lucia Nanic, Mario Laganovic, Neda

Slade, Bojan Jelakovic, and Ivica Rubelj. 2018. “Mechanisms of fetal epigenetics

32

that determine telomere dynamics and health span in adulthood.” Mechanisms of Ageing

and Development, 174: 55 –62.

Rege, Mari, Kjetil Telle, and Mark Votruba. 2011. “Parental Job Loss and Children’s

School Performance.” Review of Economic Studies, 78(4): 1462–1489.

Regmi, Krishna. 2019. “Examining the Externality of Unemployment Insurance on Chil-

dren’s Educational Achievement.” Economic Inquiry, 57(1): 172–187.

Ruhm, Christopher J. 2004. “Parental Employment and Child Cognitive Development.”

The Journal of Human Resources, 39(1): 155–192.

Schaller, Jessamyn. 2016. “Booms, Busts, and Fertility: Testing the Becker Model Using

Gender-Specific Labor Demand.” Journal of Human Resources, 51(1): 1–29.

Stevens, Ann Huff. 1997. “Persistent Effects of Job Displacement: The Importance of

Multiple Job Losses.” Journal of Labor Economics, 15(1): 165–88.

Tierney, Adrienne L., and Charles A. Nelson. 2009. “Brain Development and the Role

of Experience in the Early Years.” Zero to Three, 30(2): 9–13.

Todd, Petra E., and Kenneth I. Wolpin. 2003. Economic Journal, 113(485).

Todd, Petra E., and Kenneth I. Wolpin. 2007. Journal of Human Capital, 1(1): 91–136.

UNICEF. 2014. “Early Childhood Development. The Key to a full and productive life.”

https://www.unicef.org/dprk/ecd.pdf. Accessed 12 Novemeber, 2017.

van den Berg, Gerard J., Dorly J. H. Deeg, Maarten Lindeboom, and France

Portrait. 2010. “The Role of Early-Life Conditions in the Cognitive Decline due to Ad-

verse Events Later in Life.” Economic Journal, 120(548): 411–428.

33

van den Berg, Gerard J., Maarten Lindeboom, and France Portrait. 2006. “Eco-

nomic Conditions Early in Life and Individual Mortality.” American Economic Review,

96(1): 290–302.

Waldfogel, Jane. 2002. “Child care, women’s employment, and child outcomes.” Journal

of Population Economics, 15(3): 527–548.

Wang, Min-Chiang, and John van Ryzin. 1981. “A Class of Smooth Estimators for

Discrete Distributions.” Biometrika, 68: 301–309.

Woodhead, Martin. 2006. “Changing perspectives on early childhood: Theory, research

and policy.” Paper commissioned for the Education for All Global Monitoring Report

2007, Strong foundations: Early childhood care and education.

34

Figure 1: Distribution of Birth Cohorts in NLSY79 Child and Young Adult

Birth Year

Den

sity

1970 1980 1990 2000 2010

0.00

0.01

0.02

0.03

0.04

0.05

0.06

Note: The kernel density estimate is obtained via the Wang and van Ryzin (1981) kernelfunction for ordered discrete data with plug-in bandwidth, see Chu, Henderson and Parmeter(2015).

35

Tab

le1:

Sum

mary

Sta

tist

ics

NM

ean

Std

.D

ev.

Min

imum

Maxim

um

PIA

TM

ath

1851

90.

306

0.96

0-2

.497

2.34

8P

IAT

Rea

din

g18

460

0.22

40.

957

-2.6

412.

043

PP

VT

7576

0.45

50.

874

-3.4

583.

371

Age

1860

29.

655

2.75

25

14F

emal

e18

602

0.49

00.

500

01

Whit

e18

602

0.92

90.

257

01

His

pan

ic18

602

0.02

20.

145

01

Bla

ck18

602

0.05

00.

217

01

Mot

her

’sA

ge18

602

36.8

986.

231

056

Mot

her

’sA

FQ

T18

602

52.1

1228

.342

010

0M

other

sw

ith

som

eC

olle

ge18

602

0.23

10.

422

01

Mot

her

sw

ith

Col

lege

Deg

ree

1860

20.

249

0.43

30

1M

other

’sM

arit

alSta

tus

atC

hild’s

Bir

th18

602

0.76

70.

423

01

Num

ber

ofSib

lings

1860

20.

020

0.17

50

8P

redic

ted

Em

p.

Gro

wth

Rat

e18

602

1.13

70.

984

-1.5

034.

095

Fem

ale-

Sp

ecifi

cP

redic

ted

Em

p.

Gro

wth

Rat

e18

602

1.30

40.

887

-1.6

013.

876

Mal

e-Sp

ecifi

cP

redic

ted

Em

p.

Gro

wth

Rat

e18

602

1.00

81.

131

-2.8

764.

515

Unem

p.

Rat

e18

602

6.81

52.

215

2.30

017

.792

Note

s:S

um

mary

stat

isti

csare

calc

ula

ted

usi

ng

surv

eyw

eigh

ts.

Th

esa

mp

lein

clu

des

chil

dre

nag

es5

to14

wh

oh

ave

eith

ern

on-m

issi

ng

PIA

Tm

ath

scor

eor

PIA

Tre

adin

gsc

ore.

We

nor

mal

ize

PIA

Tm

ath

,P

IAT

read

ing,

and

PP

VT

score

s.

36

Table 2: The Effects of the Unemployment Rate at Birth on Chil-dren’s Cognitive Achievements

Math ReadingUnemp. Rate 0.002 0.009

(0.005) (0.006)Age 0.179*** -0.086***

(0.020) (0.021)Age Squared -0.009*** 0.004***

(0.001) (0.001)Female -0.070*** 0.144***

(0.025) (0.026)Hispanic -0.263** -0.130*

(0.100) (0.077)Black -0.333*** -0.233***

(0.040) (0.048)Mother’s Age 0.008 0.013**

(0.006) (0.006)Mother’s AFQT 0.008*** 0.008***

(0.001) (0.001)Mother’s Wage 0.008 0.004

(0.013) (0.011)Mother with some College 0.064* 0.102**

(0.033) (0.040)Mother with College Degree 0.278*** 0.235***

(0.051) (0.039)Mother’s Marital Status 0.110*** 0.119***

(0.018) (0.032)Number of Siblings -0.015 -0.048

(0.037) (0.037)Observations 18,519 18,460R2 0.218 0.190State-of-Birth Fixed Effects Yes YesState Fixed Effects Yes YesYear Fixed Effects Yes Yes

Notes: The dependent variables are the PIAT math and reading scores of chil-dren aged 5 to 14 years. We normalize the scores. Standard errors are clusteredat the state-of-birth level. * denotes significance at the ten percent level, **denotes at the five percent level, and *** denotes at the one percent level.

37

Table 3: The Effects of Predicted Employment Growth at Birthon Children’s Cognitive Achievements

Math ReadingPredicted Emp. Growth 0.019* 0.026**

(0.011) (0.011)Observations 18,519 18,460R2 0.218 0.190Controls Yes YesState-of-Birth Fixed Effects Yes YesState Fixed Effects Yes YesYear Fixed Effects Yes Yes

Notes: The dependent variables are the PIAT math and reading scores ofchildren aged 5 to 14 years. We normalize the scores. Standard errorsare clustered at the state-of-birth level. * denotes significance at the tenpercent level, ** denotes at the five percent level, and *** denotes at theone percent level.

Table 4: The Effects of Gender-Specific Predicted EmploymentGrowths at Birth on Children’s Cognitive Achievements

Math Reading Math ReadingPredicted Male Emp. Growth 0.018** 0.025***

(0.009) (0.009)Predicted Female Emp. Growth 0.015 0.019

(0.015) (0.015)Observations 18,519 18,460 18,519 18,460R2 0.218 0.190 0.218 0.190Controls Yes Yes Yes YesState-of-Birth Fixed Effects Yes Yes Yes YesState Fixed Effects Yes Yes Yes YesYear Fixed Effects Yes Yes Yes Yes

Notes: The dependent variables are the PIAT math and reading scores ofchildren aged 5 to 14 years. We normalize the scores. Each column representsresults from a separate regression. Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level, ** denotes atthe five percent level, and *** denotes at the one percent level.

38

Table 5: The Effects of Predicted Male Employment Growth at Birthon Children’s Cognitive Achievements

Unmarried Mothers Married Mothers

Math Reading Math ReadingPanel APredicted Male Emp. Growth -0.011 0.017 0.033*** 0.032**

(0.018) (0.025) (0.010) (0.012)Observations 4,404 4,392 14,115 14,068R2 0.245 0.283 0.198 0.164Controls Yes Yes Yes YesState-of-Birth Fixed Effects Yes Yes Yes YesState Fixed Effects Yes Yes Yes YesYear Fixed Effects Yes Yes Yes Yes

Panel BPredicted Female Emp. Growth 0.007 0.020 0.024 0.024


Notes: The dependent variables are the PIAT math and reading scores of childrenaged 5 to 14 years. We normalize the scores. Each column represents results froma separate regression. Standard errors are clustered at the state-of-birth level. *denotes significance at the ten percent level, ** denotes at the five percent level,and *** denotes at the one percent level.

39

Table 6: The Effects of Gender-Specific Predicted EmploymentGrowths at Birth on Children’s Cognitive Achievements: An Al-ternative Outcome

PPVT PPVTPredicted Male Emp. Growth 0.023*

(0.012)Predicted Female Emp. Growth 0.017

(0.045)Observations 7,853 10,409R2 0.192 0.230Controls Yes YesState-of-Birth Fixed Effects Yes YesState Fixed Effects Yes YesYear Fixed Effects Yes Yes

Notes: The dependent variable is the Peabody Picture Vocabulary Test(PPVT) score of children aged 5 to 14 years. We normalize the scores. Eachcolumn represents results from a separate regression. Standard errors are clus-tered at the state-of-birth level. * denotes significance at the ten percent level,** denotes at the five percent level, and *** denotes at the one percent level.

Table 7: The Effects of Male-Specific Employment Growth atBirth on Home Environment

HOME Score Emotional CognitionPredicted Male Emp. Growth 0.092** 0.101** 0.051

(0.036) (0.047) (0.044)Observations 1,099 1,099 1,082R2 0.445 0.365 0.413Controls Yes Yes YesState-of-Birth Fixed Effects Yes Yes YesState Fixed Effects Yes Yes YesYear Fixed Effects Yes Yes Yes

Notes: We estimate the effects of male-specific predicted employmentgrowth on three measures of home environment separately. See the maintext for detail. * denotes significance at the ten percent level, ** denotesat the five percent level, and *** denotes at the one percent level.

40

Table 8: Heterogenous Effects by Age

Aged 5-10 Aged 11-14

Math Reading Math ReadingPredicted Male Emp. Growth 0.035*** 0.027** 0.032** 0.039**



41

Tab

le9:

Hete

rogenous

Eff

ect

sby

Race

Whit

eB

lack

His

pan

ic

Mat

hR

eadin

gM

ath

Rea

din

gM

ath

Rea

din

gP

redic

ted

Mal

eE

mp.

Gro

wth

0.03

4***

0.03

4**

-0.0

11-0

.035

0.02

00.

049

(0.0

10)

(0.0

13)

(0.0

73)

(0.0

56)

(0.0

25)

(0.0

46)

Obse

rvat

ions

12,3

3412

,291

875

874

906

903

R2

0.19

00.

160

0.34

30.

382

0.34

20.

340

Con

trol

sY

esY

esY

esY

esY

esY

esSta

te-o

f-B

irth

Fix

edE

ffec

tsY

esY

esY

esY

esY

esY

esSta

teF

ixed

Eff

ects

Yes

Yes

Yes

Yes

Yes

Yes

Yea

rF

ixed

Eff

ects

Yes

Yes

Yes

Yes

Yes

Yes

Note

s:T

he

dep

end

ent

vari

able

sar

eth

eP

IAT

mat

han

dre

adin

gsc

ores

ofch

ild

ren

aged

5to

14ye

ars.

We

norm

aliz

eth

esc

ores

.E

ach

colu

mn

rep

rese

nts

resu

lts

from

ase

par

ate

regr

essi

on.

Sta

nd

ard

erro

rsar

ecl

ust

ered

atth

est

ate-

of-b

irth

leve

l.*

den

otes

sign

ifica

nce

atth

ete

np

erce

nt

leve

l,**

den

ote

sat

the

five

per

cent

leve

l,an

d**

*d

enot

esat

the

one

per

cent

leve

l.

42

Table 10: The Effects of Predicted Male Employment Growthat Birth on Children’s Cognitive Achievements: Controlling forShocks Later in Life

Math ReadingPredicted Male Emp. Growth 0.033*** 0.030**

(0.010) (0.012)Observations 14,115 14,068R2 0.198 0.168Controls Yes YesControls for Subsequent Emp. Growth Rates Yes YesState-of-Birth Fixed Effects Yes YesState Fixed Effects Yes YesYear Fixed Effects Yes Yes

Notes: The dependent variables are the PIAT math and reading scores ofchildren aged 5 to 14 years. We normalize the scores. Each column representsresults from a separate regression. Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level, ** denotes at thefive percent level, and *** denotes at the one percent level.

Table 11: Mother Fixed Effects

Math ReadingPredicted Male Emp. Growth 0.033** 0.043***


Notes: The dependent variables are the PIAT math and reading scores ofchildren aged 5 to 14 years. We normalize the scores. Each column representsresults from a separate regression. Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level, ** denotes at thefive percent level, and *** denotes at the one percent level.

43

Appendix AFigure A1: Relationship between the Unemployment Rate and Predicted EmploymentGrowths

0

5

10

1979 1984 1991 1998 2005 2009Year

Un. Rate Emp. GrowthMale. Emp. Growth Fem. Emp Growth

Note: The figure plots predicted employment growth, predicted male-specific employmentgrowth, predicted female-specific employment growth, and the unemployment rate.

44

Figure A2: Effects on Math and Reading Scores

-.05

0

.05

.1

Est

iam

ted

Coe

ffici

ents

-3 -2 -1 0 1 2 3Years Relative to Birth Year

(a)

-.1

-.05

0

.05

Est

iam

ted

Coe

ffici

ents

-3 -2 -1 0 1 2 3Years Relative to Birth Year

(b)

Notes: We estimate the effect of male-specific employment growth at ages -3, -2, -1, 0, 1, 2,and 3 on the children’s PIAT math and reading scores. Figure (a) plots the coefficients andassociated confidence intervals for math scores and Figure (b) for reading scores. 95 percentconfidence intervals are derived clustering standard errors at the state of birth level.

45

Table A1: Effects on Parents’ Labor Market Outcomes

No. of Weeks Wages No. of Weeks WagesPredicted Male Emp. Growth 0.425* 0.035**

(0.227) (0.017)Predicted Female Emp. Growth 1.530** 0.011

(0.577) (0.055)Observations 3,200 2,717 4,017 2,447R2 0.071 0.234 0.215 0.321Controls Yes Yes Yes YesState Fixed Effects Yes Yes Yes YesYear Fixed Effects Yes Yes Yes Yes

Notes: We estimate the effects of gender-specific employment growths on fathers’ and mothers’ labormarket outcomes in the year of birth. Each column represents results from a separate regression.Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level,** denotes at the five percent level, and *** denotes at the one percent level.

46

Table A2: Effects on Family Poverty Status

Family Poverty Family PovertyPredicted Male Emp. Growth -0.014**

(0.007)Predicted Female Emp. Growth -0.011

(0.013)Observations 3,478 3,478R2 0.299 0.298Controls Yes YesState Fixed Effects Yes YesYear Fixed Effects Yes Yes

Notes: The dependent variable is whether a family is living in poverty or not.Each column represents the results from a separate regression. Standard errorsare clustered at the state-of-birth level. * denotes significance at the ten percentlevel, ** denotes at the five percent level, and *** denotes at the one percentlevel.

47

Table A3: Correlations between the Unemployment Rate and Predicted Employ-ment Growths at Birth

Unemp. Rate Unemp. Rate Unemp. RatePredicted Emp. Growth -0.415***

(0.079)Predicted Male Emp. Growth -0.363***

(0.055)Predicted Female Emp. Growth -0.202**

(0.081)Observations 1,581 1,581 1,581R2 0.766 0.768 0.760State Fixed Effects Yes Yes YesYear Fixed Effects Yes Yes Yes

Notes: We analyze correlations between the unemployment rate and predicted employmentgrowths. We use state-level data from 1979 to 2009, the birth period of children in our mainanalysis. Standard errors are clustered at the state-of-birth level. * denotes significance at theten percent level, ** denotes at the five percent level, and *** denotes at the one percent level.

48

Table A4: Using both Male- and Female-Specific EmploymentGrowths

Math ReadingPredicted Male Emp. Growth 0.048** 0.046**

(0.020) (0.022)Predicted Female Emp. Growth -0.024 -0.022


Notes: The dependent variables are the PIAT math and reading scores of childrenaged 5 to 14 years. We normalize the scores. We use both male- and female-specific employment growths in the same regression. Each column representsresults from a separate regression. Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level, ** denotes at the fivepercent level, and *** denotes at the one percent level.

49

Table A5: The Effects of Predicted Male Employment Growth atBirth on Children’s Cognitive Achievements: Limiting to Chil-dren Born Before 2001

Math ReadingPredicted Male Emp. Growth 0.025*** 0.029**



Table A6: The Effects of Predicted Male Employment Growthat Birth on Children’s Cognitive Achievements: Using ChildrenWho Were Born Before July

Math ReadingPredicted Male Emp. Growth 0.063*** 0.062***



50

Tab

leA

7:A

ddit

ional

Robust

ness

Check

s

Model

(1)

Model

(2)

Mat

hR

eadin

gM

ath

Rea

din

gP

redic

ted

Mal

eE

mp.

Gro

wth

0.03

6***

0.02

8**

0.03

4***

0.03

1**

(0.0

11)

(0.0

11)

(0.0

10)

(0.0

12)

Obse

rvat

ions

14,0

1713

,969

14,1

1514

,068

R2

0.19

80.

164

0.19

20.

157

Con

trol

sY

esY

esY

esY

esSta

te-o

f-B

irth

Fix

edE

ffec

tsY

esY

esN

oN

oSta

teF

ixed

Eff

ects

Yes

Yes

Yes

Yes

Yea

rF

ixed

Eff

ects

Yes

Yes

Yes

Yes

Note

s:T

he

dep

end

ent

vari

able

sar

eth

eP

IAT

mat

han

dre

adin

gsc

ores

ofch

ild

ren

aged

5to

14ye

ars.

We

nor

mal

ize

the

scor

es.

Eac

hco

lum

nre

pre

sents

the

resu

lts

from

ase

par

ate

regr

essi

on.

Mod

el(1

)fu

rth

erad

ds

the

lagg

edm

ale-

spec

ific

emp

loym

ent

grow

thin

our

mai

nsp

ecifi

cati

onas

aco

ntr

olva

riab

le.

Mod

el(2

)ad

ds

ind

ust

rysh

ares

inth

eb

ase

year

inou

rb

asel

ine

spec

ifica

tion

.S

tan

dar

der

rors

are

clu

ster

edat

the

stat

e-of

-bir

thle

vel.

*d

enot

essi

gnifi

can

ceat

the

ten

per

cent

leve

l,**

den

otes

atth

efi

vep

erce

nt

leve

l,an

d**

*d

enot

esat

the

one

per

cent

level

.

51

Table A8: The Effects of the Unemployment Rate on Children’s Cog-nitive Achievements: An Instrumental Variable Approach

Math ReadingUnemp. Rate -0.040*** -0.039***

(0.012) (0.014)First Stage F-Statistic 384.62 399.59Observations 14,115 14,068Controls Yes YesState-of-Birth Fixed Effects Yes YesState Fixed Effects Yes YesYear Fixed Effects Yes Yes

Notes: We estimate the effect of the unemployment rate in the year of child’sbirth on children’s PIAT math and reading scores, using predicted employmentgrowth as an instrumental variable. The scores are normalized. Each columnrepresents the results from a separate regression. Standard errors are clusteredat the state-of-birth level. * denotes significance at the ten percent level, **denotes at the five percent level, and *** denotes at the one percent level.

52

Table A9: Effects on Fertility

Give Birth Give BirthPred. Male Emp. Growth*Coll. Deg. -0.002

(0.002)Pred. Male Emp. Growth 0.001

(0.003)Pred. Female Emp. Growth*Coll. Deg. 0.000

(0.003)Pred. Female Emp. Growth 0.006

(0.005)Coll. Deg 0.018*** 0.017***

-0.003 -0.003Observations 36,448 36,448R2 0.066 0.066Controls Yes YesState Fixed Effects Yes YesYear Fixed Effects Yes Yes

Notes: The dependent variable is whether or not a married mother gave birth inthe survey year. If a mother reports having a child under the age of one year, weinterpret this as mother giving birth in the survey year. Each column representsresults from a separate regression. Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level, ** denotes at the fivepercent level, and *** denotes at the one percent level.

53

Appendix B: Not for Publication

Alternative Measures of Outcomes. In our main analysis, we use a child-year as a unit of

analysis, that is, we allow the labor market shock at birth to have effects on children’s

cognitive growth at different points of age during childhood. In this section, we consider

having only one outcome for each child. To do so, we use two different approaches. First,

we collapse test scores at the child level. Second, we use the first score of each child that we

observe in our sample. As reported in Table B1, our results are consistent with those of the

main sample.

Without Controls for Mother Characteristics. We extend our analysis by estimating the

effect of predicted female employment growth without controlling for mother characteristics.

We show in the main text that an improvement in women’s employment opportunities has

an insignificant effect on children’s cognitive development. We further explore here if our

estimates change when we do not control for mother characteristics. We use the specification

parallel to Equation 5, except that we do not control for mother characteristics. The results,

which are presented in Table B2, are consistent with our baseline estimates.

Additional Robustness Checks. To account for the fact that children are observed multiple

times in our sample, we cluster our standard errors both at the state-of-birth and at the

child level (Table B3). Table B4 presents the results estimated using state-specific linear

trends.

Race-Specific Predicted Male Employment Growths. We construct the race-specific predicted

male employment growths, using the following approach:

Lmale,rst =17∑k=1

$kmale,rs,1970∆N

kt (7)

where r ∈ {white, black, hispanic} and $kmale,rs,1970 is the ratio of each race male employment

in an industry to the total male employment of each race in a state in the base year. We

54

estimate the model analogues to Equation 5 in the main text. As shown in Table B5, the re-

sults are similar to those results derived from applying the general male-specific employment

growth rate in the main text.

55

Table B1: Alternative Measures

Model (1) Model (2)

Math Reading Math ReadingPredicted Male Emp. Growth 0.031*** 0.029** 0.088*** 0.036*

(0.010) (0.014) (0.021) (0.019)Observations 3,466 3,465 3,454 3,404R-squared 0.263 0.219 0.183 0.230Controls Yes Yes Yes YesState-of-Birth Fixed Effects Yes Yes Yes Yes

Notes: In model (1), we average children’s PIAT math and reading scoresand use averages as dependent variables. Model (2) uses test scores that areobserved for the first time for each child. Standard errors are clustered at thestate-of-birth level. * denotes significance at the ten percent level, ** denotesat the five percent level, and *** denotes at the one percent level.

56

Table B2: The Effect of the Predicted Female Employment GrowthRate on Children’s Cognitive Achievements: Without Controllingfor Mother Characteristics

Math ReadingPredicted Female Emp. Growth 0.016 0.021


Notes: The dependent variables are the PIAT math and reading scores of chil-dren aged 5 to 14 years. We normalize the scores. Each column representsresults from a separate regression. Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level, ** denotes at the fivepercent level, and *** denotes at the one percent level.

57

Table B3: Clustering Standard Errors at both the State-of-Birth andChild Level



Notes: The dependent variables are the PIAT math and reading scores of chil-dren aged 5 to 14 years. We normalize the scores. Each column representsresults from a separate regression. Standard errors are clustered at both thestate-of-birth and child level. * denotes significance at the ten percent level, **denotes at the five percent level, and *** denotes at the one percent level.

58

Table B4: Controlling for State-Specific Linear Trend



Notes: The dependent variables are the PIAT math and reading scores of chil-dren aged 5 to 14 years. We normalize the scores. Each column representsresults from a separate regression. Standard errors are clustered at the state-of-birth level. * denotes significance at the ten percent level, ** denotes at the fivepercent level, and *** denotes at the one percent level.

59

Tab

leB

5:T

he

Eff

ect

of

Race

-Sp

eci

fic

Pre

dic

ted

Male

Em

plo

ym

ent

Gro

wth

on

Chil

dre

n’s

Cognit

ive

Ach

ievem

ents

Mat

hR

eadin

gM

ath

Rea

din

gM

ath

Rea

din

g

Pre

dic

ted

Em

p.

Gro

wth

for

Whit

eM

en0.

034*

**0.

033*

*(0

.010

)(0

.013

)P

redic

ted

Em

p.

Gro

wth

for

Bla

ckM

en0.

005

-0.0

25(0

.066

)(0

.056

)P

redic

ted

Em

p.

Gro

wth

for

His

pan

icM

en0.

021

0.03

2(0

.027

)(0

.043

)O

bse

rvat

ions

12,3

3412

,291

875

874

906

903

R2

0.19

00.

160

0.34

30.

382

0.34

20.

339

Con

trol

sY

esY

esY

esY

esY

esY

esSta

te-o

f-B

irth

Fix

edE

ffec

tsY

esY

esY

esY

esY

esY

esSta

teF

ixed

Eff

ects

Yes

Yes

Yes

Yes

Yes

Yes

Yea

rF

ixed

Eff

ects

Yes

Yes

Yes

Yes

Yes

Yes

Note

s:T

he

dep

end

ent

vari

able

sar

eth

eP

IAT

mat

han

dre

adin

gsc

ores

ofch

ild

ren

aged

5to

14ye

ars.

We

nor

mali

zeth

esc

ore

s.E

ach

colu

mn

repre

sents

resu

lts

from

ase

par

ate

regr

essi

on.

Sta

nd

ard

erro

rsar

ecl

ust

ered

at

the

stat

e-of-

bir

thle

vel.

*d

enot

essi

gnifi

can

ceat

the

ten

per

cent

level

,**

den

otes

atth

efi

vep

erce

nt

level

,an

d***

den

otes

at

the

on

ep

erce

nt

level

.

60

Labor Demand Shocks at Birth and Cognitive Achievement ...€¦ · 1E.g., UNICEF 2014, Woodhead...

Documents

Transcript of Labor Demand Shocks at Birth and Cognitive Achievement ...€¦ · 1E.g., UNICEF 2014, Woodhead...